Abstract: Subsurface reservoirs commonly exhibit layered structures. Conventional amplitude variation with angle (AVA) inversion, which relies on the Zoeppritz equation and its approximations, often fails to accurately estimate elastic parameters because it assumes single-interface models and ignores multiple reflections and transmission losses. To address these limitations, this study proposes a novel prestack time-frequency domain joint inversion method that utilizes the reflection matrix method (RMM) as the forward operator. The RMM accurately simulates wave propagation in layered media, while the joint inversion framework minimizes the misfit between observed and synthetic data in both the time and frequency domains. By incorporating Bayesian theory to optimize the inversion process, the method effectively balances contributions from both time-domain waveforms and frequency-domain spectral information through a weighting factor. Tests on both synthetic data and field data demonstrate that the proposed method outperforms conventional AVA inversion and time-domain waveform inversion in accuracy and robustness. Furthermore, the method demonstrates good robustness against variations in initial models, random noise, and coherent noise interference. This study provides a practical and effective approach for high-precision reservoir characterization, with potential applications in complex layered media.